基于DQN的电力协同计算与缓存的任务卸载策略

DQN-based Task Offloading Strategy for Power Co-computing and Caching

多接入边缘计算为满足配电网海量终端接入的通信和计算需要提供了一种有效的解决方案,边缘计算通过将实时采集的终端设备数据上传到距离更近的边缘服务器,进行数据处理和存储,从而降低了通信时延和减轻了云服务器的负担。然而,由于系统环境的动态变化,任务卸载优化成为一个具有挑战性的问题。基于强化学习和协同缓存模型,提出了联合无线信道条件、边缘服务器和云服务器的带宽资源与算力资源的云-边-端协同任务卸载策略。该策略结合了深度Q网络(deep Q-network,DQN)算法与协同缓存模型,旨在满足通信时延和能耗约束的前提下,最大程度地提高配电网系统的传输效率、安全性和处理效率。仿真结果表明,相比于其他卸载策略,所提策略在最大任务数据量、终端设备数量和边缘服务器最大计算频率方面可以有效降低系统时延与能耗。

Multiaccess edge computing provides an effective solution for meeting the communication and computation requirements of massive terminal access in distribution networks.Edge computing reduces communication latency and the burden on cloud servers by uploading terminal device data collected in real-time to a closer edge server for data processing and storage.However,task offloading optimization has become a challenging problem owing to the dynamic changes in the system environment.Based on reinforcement learning and collaborative caching models,a cloud-edge-end collaborative task offloading strategy that combines wireless channel conditions,bandwidth resources,and arithmetic resources of edge and cloud servers was proposed.This strategy combines the deep Q-network(DQN)algorithm with the cooperative caching model to maximize the transmission efficiency,security,and processing efficiency of the distribution network system while satisfying the communication delay and energy consumption constraints.Simulation results reveal that compared with other offloading strategies,the proposed strategy effectively reduces system delay and energy consumption in terms of the maximum task data volume,number of end devices,and maximum computing frequency of the edge server.